Downscaling by Pseudo Global Warning Method
1
1.
Fujio KIMURA 1 and Akio KITOH2
Graduate School of Life and Environmental Sciences, University of Tsukuba
1-1-1 Tennoudai, Tsukuba, Ibaraki 305-8272, Japan
2
Meteorological Research Institute, Japan Meteorological Agency
Objectives
Climate change by increasing of greenhouse
gas is usually estimated by General Circulation
Model (GCM). However, horizontal resolution of
the ordinary GCM is quite low, i.e., grid interval is
about 100-300km, although these are being
improving much with the computer power day by
day. However, the resolution is still not enough to
estimate the climate change in a basin, such as
Seyhan river basin in Turkey. Downscaling of GCM
using Regional Climate Model (RCM) may arrow to
estimate climate and provides scenarios of the likely
climate change in a basin, although GCMs and the
methods of downscaling still have many problems
for the reliable projection.
In generally, one of the largest difficulty in the
downscale process using a nested regional climate
model, is the bias of GCMs, especially shift of a
regional scale climate system may gives serious
error in the nested model (Wang et al, 2004). To
avoid this difficulty, we present a new downscaling
method called PGWM (Pseudo Global Warming
Method) in which the boundary conditions are
assumed to be a linear coupling of the reanalysis
data (observation) and the difference component of
the global warming estimated by GCMs. This
assumption may valid when the change of the global
warming is small enough and allows to neglect the
nonlinear interaction between the climate change
and the inter-annual variation of the regional climate
systems.
2.
The difference of them is the climate change by the
increase of greenhouse gases. Some of the 6 hourly
reanalysis data and the different components, which
are provided at each grid point and each month, are
assumed to be as the boundary condition for the
RCM in the future (PGWM run). The downscaled
climate will be compare to the hindcast data.
A similar nesting method named 'anomaly
nesting' has been presented by Vasubandhu and
Kanamitsu 2004. Object of their method is to
improve seasonal forecasts by regional climate
models. They tested this method for a simulation of
regional climate affected by some large scale
climate systems such as ENSO or inter-annual
variability around South America. A big difference
from PGWM is that the boundary data are assumed
to be the GCM products whose climate values have
been replaced by the climate values estimated from
the reanalysis data. Short term components, such as
daily variation, are almost retain in GCM produces,
while they will be replaced by these of reanalysis
data in PGWM.
One of the advantage of PGWM is to allow to
Reanalysis
Data
GCM
Projection
6
Monthly
hourly mean
Current Future
Climate Climate
MRI-CGCM
Pseudo
Warming
Boundary
Method
Figure 1 shows a flow chart of downscaling
by PGWM. Reanalysis data are provided every 6
hours and are assumed as a boundary condition of
the RCM. In the RCM, the reanalysis data are
further interpolated for every time step when RCM
simulates the current climate. This process
reproduces the past climate using reanalysis data,
which is called 'hindcast'. The hindcasted data are
compared to in situ observation data and the results
give feedback to the RCM in order to tune some
parameters in the RCM. Two data sets of monthly
climate are obtained by ten-years average of GCM
produces. One is for the current climate during
1990s (period A) and another is for 2070s (period B).
Regional Climate Model
Downscaled
Histrical
Climate
Observation
Data
Downscaled
Climate
Projection
Conparison
for
RCM
Validation
Comparison for
Regional Climate Change
Fig.1. Flow chart of downscaling by pseudo global
warming method (PGWM).
M thl
P
ii J
@T k
Monthly Precipi Jul @Turkey Area
A
45
110
100
90
40
35
30
0
60
50
40
2
20
Obs (A)
5
5
Cont (A)
PGWM (B)
Monthly Precipi Jan @ Seyhan Area
110
100
90
80
Monthly Precipi Jul @ Seyhan Area
45
Drct (A)
Drct (B)
40
2
0
40
30
20
10
10
0
Fig. 3. .Projected monthly precipitation in January
(left) and July (right) during the period A (1998-2002)
and the period B (Five years in 2070s). Top panels
indicate mean values in the entire Turkey and the
bottom panels indicate those in the Seyhan basin.
White bar indicates observed data in the period A, blue
indicates the control run (hindcast in the period A), red
indicates projection by PGWM (period B), green
indicates the direct downscale from GCM (period A)
and pink indicates the direct downscale in period B.
Fig. 2. Top panel:Monthly mean precipitation of in
situ observation, January 1998-2002. Middle
panel: Hindcasted precipitation in the same area
and the same period as the top panel. Bottom:
Simulated precipitation by the direct downscaling
form the products of MRI-CGCM in the same
period. Color bar is common for three panels.
estimate the global warming effects on the specific
past year. This advantage makes easy to estimate
climate difference between current and future
climate without ensemble of numerous number of
simulations. Usually the effects is only possible to
estimate by the difference between the climatic
mean of many years before and after global
warming. Any GCM can not estimate the difference
between each single year without large effects of
natural inter-annual variation. When the difference
of the global warming is smaller, detection of the
global warming effect becomes more difficult
because of inter-annual variation. This method have
been already applied to Mongolia (Sato et al, 2006).
Comparison with direct down scaling
Downscaling by PGWM and ordinary direct
downscaling are compared in order to discuss the
accuracy and reliability of the method. Beside five
years hindcast (control run) using NCEP/NCAR
reanalysis data and five years projection during
2070s by PGWM, directly nesting runs driven by
daily GCM products are also carried out for the
periods A and B. Test periods are restricted to
January and July. Global-scale projection data were
provided by MRI-CGCM-2 (Yukimoto et al, 2001)
assuming A2 scenario in Special Report on
Emission Scenarios (SRES) (IPCC, 2000).
Figure 2 indicates monthly mean precipitation
of in situ observation, January 1998-2002 (top
panel), hindcasted precipitation in the same area and
the same period (middle panel) and simulated
precipitation during five years in 1990s by the direct
downscaling form the products of MRI-CGCM
(bottom). Color bar is common for three panels. We
chose January and July for the test of PGWM, since
amount of precipitation becomes large in winter and
small in summer in this region. The hindcasted
precipitation agree well to the observation,
particularly heavy precipitation along the Black Sea
and some areas along the Mediterranean. Horizontal
distribution of the direct downscale also agree to the
observation, but the amount is overestimated.
Projected total amount of precipitation in
January and July are shown in Figure 3 for the entire
Turkey (top) and for the finest grid system around
Seyhan basin (bottom). White bar indicates
observed data in the period A, blue indicates the
control run, namely; hindcast in the period A.
Hindcast slightly overestimated in July but slightly
underestimated in January. The red bars indicate
projected precipitation during 2070s by PGWM
(period B). The model projects that amount of
precipitation will decrease in January but it slightly
increase in July. Green bars and pink ones are five
years mean precipitation by the direct downscaling
from GCM in the period A and that in the period B,
respectively. The direct downscaling overestimates
and seriously underestimates during July.
Figure 4 indicate horizontal distribution of the
difference in monthly precipitation between period
A (1998-2002) and B (five years in 2070s) in
January projected by PGWM (top) and by the direct
downscaling (bottom). Dark blue indicate increasing
in precipitation, while brown indicate decreasing.
Precipitation change is depend on place in Turkey.
The patterns of horizontal distribution have good
similarity except for the Southeast corner of the
Black Sea. Figure 5 is the same as Fig.4, but the
finest grid system. The tendency is the same as
Fig.4. These projections agree well each other, since
the amplitude is somewhat larger in the direct
simulation.
Figure 6 indicates the inter-annual variation of
precipitation (top) and temperature (bottom).
Inter-annual variation of the direct downscaling for
the past year does not need to agree to observation,
since only statistics of the variation has meaning.
Temperature given by the direct downscaling from
GCM-2 has strong cold bias. The hindcast (CTL)
follows the inter-annual variation very well.
Differences between the current years and the
corresponding pseudo global warming year depend
on years, but the amplitude of inter-annual variation
of the difference is much smaller than those between
a single year of period A and that of period B
Fig.4. Difference in monthly precipitation between
period A (1998-2002) and B (Five years in 2070s) in
January. Top: projected by PGWM, Bottom: by the
direct downscaling from GCM. Dark blue indicate
increasing in precipitation, while brown indicate
decreasing.
3.
Conclusions
PGWM not only reduces large scale model bias,
it allows to estimate climate difference between
current and future climate without ensemble of
numerous number of simulations. This method has
the certain advantages, but it needs further study to
make sure the reliability for the extreme events.
Fig. 5. Same as Fig.4, but for the finest grid system
around Seyhan basin.
160
MnthlyPercipi. Jan @TurkeyArea
DMI
140
CTL
120
MRI-Pse
100
GCMr1
GCMr2
80
60
40
20
0
1998
28
1999
2000
2001
2002
ave
MnthlyTemp. Jul @TurkeyArea
DMI
27
CTL
26
MRI-Pse
25
GCMr1
GCMr2
24
23
22
21
20
1998
1999
2000
2001
2002
ave
Fig. 6. Year to year variation of precipitation (top) and
temperature (bottom). Meaning of color is same as
Fig.3.
4.
Reference
Sato, T., F. Kimura and A. Kitoh, 2006: Projection
of global warming onto regional precipitation
over Mongolia using a regional climate model,
accepted by Journal of Hydrology.
Vasubandhu, M. and M. Kanamitsu, 2004: Anomaly
Nesting: A Methodology to Downscale
Seasonal Climate Simulations from AGCMs.
Journal of Climate, vol. 17, Issue 17,
pp.3249-3262
Wang, Y., L. R. Leung, J. L. McGregor, D. -K. Lee,
W. -C. Wang, Y. Ding, and F. Kimura, 2004:
Regional
climate
modeling:
Progress,
Challenges, and Prospects. J. Meteor. Soc.
Japan., 82, 1599-1628.
Yukimoto, S., A. Noda, A. Kitoh, M. Sugi, Y.
Kitamura, M. Hosaka, K. Shibata, S. Maeda
and T. Uchiyama, 2001: A new Meteorological
Research
Institute
coupled
GCM
(MRI-CGCM2) - its climate and variability -.
Pap. Met. Geophys., 51, 47-88.